NBER WORKING PAPER SERIES

LAND AND POWER: THEORY ANDEVIDENCE FROM CHILE

Jean-Marie BalandJames A. Robinson

Working Paper 12517http://www.nber.org/papers/w12517

NATIONAL BUREAU OF ECONOMIC RESEARCH1050 Massachusetts Avenue

Cambridge, MA 02138September 2006

We are grateful to Mauricio Benitez-Iturbe and Sebastian Mazzuca for their outstanding research assistance.Our greatest debt is to Jonathan Conning who persevered through at least three previous versions. We havealso benefitted from the suggestions and advice of Daron Acemoglu, Siwan Anderson, Arnold Bauer, SamuelBowles, Michael Carter, Eduardo Fajnzylberg, Jeffry Frieden, Tu Jarvis, Sebastian Laurent, FrancoisManiquet, Dina Mesbah, Sripad Motiram, Andrew Newman, Thomas Piketty, Timothy Scully, WilliamSummerhill, Werner Troesken, and numerous seminar participants, particularly Timothy Besley, AlbertoDiaz-Cayeros, Esther Duflo, Jim Fearon, Dilip Mookherjee, Jeffrey Nugent, T.N. Srinivasan and BarryWeingast. Jean-Marie Baland would like to thank the MacArthur network on Inequality and EconomicPerformance, the CRED for financial support and the Belgian Program on Inter-University Poles ofAttraction initiated by the Belgian State, Prime Minister's office, Science Policy Programming. The viewsexpressed herein are those of the author(s) and do not necessarily reflect the views of the National Bureauof Economic Research.

©2006 by Jean-Marie Baland and James A. Robinson. All rights reserved. Short sections of text, not toexceed two paragraphs, may be quoted without explicit permission provided that full credit, including ©notice, is given to the source.

Land and Power: Theory and Evidence from ChileJean-Marie Baland and James A. RobinsonNBER Working Paper No. 12517September 2006JEL No. D72, O54, Q15

ABSTRACT

We study the connection between employment and political control. Many employment relationshipsconcede rents to workers. For example, when worker effort is crucial for production, but onlyimperfectly observed. We show that, depending on the political institutions, the presence of suchrents allows employers to use the threat of withdrawing them to control their workers' politicalbehavior. We thus demonstrate that employment does not simply generate income, it also givespower to control the behavior of others. The analysis focuses on the salient example of politicalcontrol, where landlords coerce the votes of their workers in the absence of a secret ballot. Themodel we develop generates predictions about electoral outcomes which can be tested byinvestigating the impact of the introduction of an effective secret ballot. Such an institutional reformreduces landlords' control, and in consequence, we should observe changes in voting behavior, sinceworkers whose votes were previously controlled and sold can now vote freely. We test thepredictions of the model by examining in detail the effects of the introduction of the secret ballot inChile in 1958. We show that, consistent with our theory, the political reforms led to large changesin voting behavior. Before the reforms, localities with more pervasive patron-client relationshipstend to exhibit a much stronger support for the right-wing parties, traditionally associated with thelanded oligarchy. After the reform however, this difference across localities completely disappeared.

Jean-Marie BalandCred, Department of EconomicsFUNDP8, rempart de la viergeB5000 NamurBELGIUMjean-marie.baland@fundp.ac.be

James A. RobinsonHarvard UniversityDepartment of GovernmentN309, 1737 Cambridge StreetCambridge, MA 02138and NBERjrobinson@gov.harvard.edu

\It is the most cruel mockery to tell a man he may vote for A or B,

when you know that he is so much under the inuence of A, or the

friends of A, that his voting for B would be attended with the destruc-

tion of him. It is not he who has the vote, really and substantially, but

his landlord, for it is for his bene�t and interest that it is exercised in

the present system." David Ricardo ([1824], 1951-1973, p. 506)

1. Introduction

In this paper we study the connection between employment and political control.

Many employment relationships concede rents to workers. For example, when

worker e�ort is crucial for production, but only imperfectly observed. We show

that the fact that employers already concede rents to workers may allow them,

depending on the political institutions, to use the threat of withdrawing these

rents to control their political behavior. We thus demonstrate that employment

does not simply generate income, it also gives power to control the behavior of

others.

The most salient example of such a connection arises in the absence of a secret

ballot. When voting is not secret, it becomes feasible to buy, sell and coerce

votes. While there are recorded instances of an individualized market for votes,

the main stylized fact which emerges from the case study literature is that rather

than individuals freely selling their votes to politicians, employers usually control

and supply the votes of their employees in exchange for money, favors or policies.

More speci�cally, as discussed by Ricardo (1824), employers are usually landlords.

That landlords control the political activities of their workers has historically

been a pervasive characteristic of agrarian economies.1 In Britain, before the

introduction of the secret ballot in 1872, this factor was critical in determining

the outcome of rural elections. As observed by Lord Stanley in 1841,

\when any man attempted to estimate the probable result of a county

election in England, it was ascertained by calculating the number of

1As Malefakis (1970, p. 98) summarized the situation in nineteenth century rural Andaluc��a,\a man's job depended on his vote."

3

the great landed proprietors in the county and weighing the number

of occupiers under them."2

Throughout the nineteenth century radicals and reformers complained about

the lack of a secret ballot in Britain (see Kinzer, 1982, Cox, 1987 and Gash, 1977).

In Germany, despite the fact that a democratic parliament was introduced in 1848

there is a mass of evidence that rural voters were controlled by landed interests.

Bismarck even supported an extension of voting rights in 1871 because he thought

that the control exercised by landlords over rural voters would o�set the rising

inuence of urban workers (Bendix's, 1964, p. 97, Hamerow 1974, pp. 299-300).

Landlords control over rural elections was greatly facilitated where balloting

was open (see Goldstein, 1983, p. 15). However, even when there was a supposedly

secret ballot (and not open voting), strategies were found to keep voting under

control. Thus, in the German case, political parties often printed their own ballots:

\given that ballots had to be obtained from the candidates themselves or from

their agents, it was often physically impossible for a poor man to vote for anyone

but the squire's choice." (Anderson, 1993, p. 1467)3 Even countries, such as

France which moved early to universal male su�rage (after 1848) and free elections

(after 1871) only introduced an e�ective (though non-Australian)4 secret ballot in

1913. Before this \the ballots frequently had subtle but distinct marks, such as

paper thickness, colour and size, from which the election o�cials could deduce a

voter's decision. This information was then passed on to notables who could easily

punish such wayward voters since they frequently were his tenants or employees,"

(Kreuzer 1996, p. 108).

Similar tactics were used and remain up to the present day in democratic third

2Quoted in Kitson-Clark (1951, p. 112). O'Gorman (1989, p.20) estimates that by 1807 thisresulted in the outcomes of 300 parliamentary seats being a foregone conclusion. He describesin detail the system of patronage linking high politicians such as Walpole with members ofparliament, typically Whig `oligarchs,' who controlled the local electorate.

3For further evidence on the German case see Blackbourn (1988) and the section on Germanyin Gibson and Blinkhorn (1991).

4The `Australian ballot' has become synonymous with perfectly secret voting and refers toa situation where all political alternatives are on a single government produced ballot paper. Itderives its name from the fact that the �rst use of such a ballot was in Australia in 1856.

4

world countries.5 Nowhere is the evidence about landlord control of elections so

conclusive as in Latin America. Following independence most Latin American

countries adopted liberal constitutions committing themselves to regular elections,

yet, with few exceptions Latin American societies did not become consolidated

democracies with free regular elections contested by all adults until the 1980's.6

In Colombia, the country which has the longest experience of formal democracy in

Latin America and where the military has played the most marginal of political

roles, the equivalent of the Australian ballot was legislated only in 1988 and

introduced �rst in 1990. In Chile the control of voting by landowners was very

frankly discussed in the debate leading up to the introduction of the secret ballot

in 1958 in language strikingly similar to that used by Lord Stanley quoted above.

For example, Socialist senator Martones argued in favor of introducing the secret

ballot because,

\if that law [the old electoral law without a secret ballot] did not exist,

instead of there being 9 Socialist senators there would be 18, and you

[the Conservatives] would be reduced to 2 or 3 ... [laughter] you laugh,

but the truth is that there would be not 2 Conservative senators from

O'Higgins and Colchagua, which corresponds exactly to the number of

inquilinos in the fundos which belong to the Conservative hacendados

in that region. Conservatives would have only one or perhaps none."7

We develop a model of labor contracting with moral hazard and limited liability

in the rural sector.8 Absent politics, landlords �nd it optimal to concede rents

5For evidence from India see Kohli (1990, pp. 227-228) and Breman (1974).6See Engerman and Sokolo� (2005) and Hartlyn and Valenzuela (1998). For Brazil see Pang

(1973), Graham (1990) and Martins (1996).7A \fundo" is a large farm and a \hacendado" a large landowner and an inquilino was a

permanent worker on such farms. Quoted in El Mercurio, Saturday May 19, 1958, p. 20.8Why does electoral corruption seem to be more signi�cant in rural as opposed to urban areas?

Our model suggests three factors which indicate that urban votes may be more expensive to buyand hence political corruption relatively less attractive in urban settings. First, as suggestedby Marglin (1973) it may be the case that rents are lower in factories and urban environments.This means that, to induce e�ort, employers need to pay only a small wage premium. Second,it may be harder to monitor voting and political activities in anonymous urban environments.

5

to workers to induce e�ort. Introducing politics, we show that the presence of

these rents allows landlords to pay less than the full opportunity cost of the votes

of their workers, and thus possibly to pro�tably o�er a contract stipulating both

economic and political (voting) behavior.9 This feature also implies that it is

cheaper for political parties to buy votes indirectly through landlords since this

means they can avoid fully compensating workers for the value of their votes. We

study the implications of these phenomena for the functioning of factor markets.

The model we develop generates predictions about electoral outcomes which

can be tested by investigating the impact of the introduction of an e�ective secret

ballot. Such an institutional reform reduces landlords' control, and in conse-

quence, we should observe changes in voting behavior, since workers whose votes

were previously controlled and sold can now vote freely.

We examine these implications by considering the introduction of the secret

ballot in 1958 in Chile. We show in section 4 that the empirical predictions of

our model are highly consistent with the data. We �rst demonstrate that before

1958 control of voting was endemic in the traditional `oligarchic' provinces of the

so-called North and Urban Central Valley provinces which were characterized by

long term patron-client relationships known in Chile as the inquilinaje system (see

e.g. Sadoulet, 1992). More precisely, we show that, before the reform, the support

for right-wing parties was substantially higher, but that it also fell substantially

more thereafter in precisely those municipalities where inquilinos formed a larger

share of the electorate.

The case study literature on the secret ballot focuses very much on coercion

and corruption and has obviously noticed the fact that systematic biases can be

introduced into elections because of such phenomena (e.g., our discussion of nine-

teenth century Germany above). In the Chilean case, scholars such as Loveman

(1976), Scully (1992) or Bauer (1995), have noted the signi�cance of the control

of inquilinos for the political power of the right before 1958, and linked the intro-

Third, workers in the cities may enjoy superior alternative employment opportunities, whichreduce the scope for political control by their employers.

9Though we model political control in terms of voting, the analysis extends to other types ofpolitical activities (such as types of protests, riots, demonstrations and other forms of collectiveaction). Such a model would have similar results to the one we present.

6

duction of the secret ballot to the rise of socialism. Our contribution here is to

provide a microfoundation for why landlords control voting behavior and provide

the �rst systematic test of the impact of the 1958 reforms in Chile.

The incidence of secret balloting has been understudied by the literature on

political economy and institutions (e.g., Cox, 1997, Persson and Tabellini, 2000,

2003, Acemoglu and Robinson, 2006). Piketty (1999, 2000) and Dekel, Jackson

and Wolinsky (2004) developed models of how an individualized market for votes

might work and have studied the circumstances under which vote buying is socially

undesirable. Related papers by Snyder (1991) and Grossman and Helpman (1996)

have looked at interest groups buying politicians with `campaign contributions.'

All of these theoretical papers focus on very di�erent issues than those we study.

An important distinction is that these scholars, and most others in the political

economy literature, focus on the e�ciency of government policy. We focus on why

vote buying is linked to employment and how this inuences factor markets. We

also provide empirical evidence supporting our model. Most closely related to our

research, Summerhill (1995) developed a simple model of the idea that political

rents accrue to landowners and tried to estimate the impact of electoral reform

on the economy using data from nineteenth century Brazil.

2. The Model

2.1. The Fundamentals

We consider a discrete time in�nite horizon model of the rural sector. There is a

unit mass of agents and a proportion x of rural agents have access to the capital

market and can therefore purchase land and hire workers. All rural agents have

the option to be self-employed and earn an income of w. We let m denote the

proportion of rural agents who become agricultural workers, and 1�m� x thosewho remain self-employed.

There are L units of land which are owned by landowners with each owning

L=x = l units of land. There is a single numeraire consumption good which is

produced from land and labor. The technology is characterized by a standard

7

constant returns to scale neoclassical production function. On a farm, output of a

worker in any period is equal to ~�g�ln

�where n is employment, g is the intensive

form of the production function so that g0 > 0 and g00 < 0, and ~� is a plot-speci�c

stochastic shock to output which is distributed independently across plots and

time and can take two values, � and 0 (by normalization) (since we focus on

stationary equilibria we do not introduce time subscripts). The probability that

� occurs in period t depends on the e�ort exerted by a worker in that period.10

E�ort, ", takes two values, " 2 f0; eg. If " = e, � occurs with probability h, whileif " = 0, � occurs with probability l < h.

While output is perfectly observable by the landlord, the level of e�ort exerted

by the worker is not. This induces a moral hazard problem. We assume that

e�ort can never be observed so that the only possible wage contract depends on

the realization of ~�.

There are also two political parties, `Left' (denoted L) and `Right' (denoted

R) competing for votes to win an election and all individuals have exogenous

preferences for one of these parties which means that they get utility from voting

for the party they prefer (as in a standard probabilistic voting model).

All agents in the rural sector have utility functions which are linear in con-

sumption, c, e�ort, ", and voting decision �j for j = L;R which depends on the

ideological orientation of the agent. Thus, U(c; "; �j) = c � " + �j is the util-ity of an agent of type j if they vote for the party they prefer, otherwise it is

U(c; "; �j) = c� ". All agents maximize the expected present discounted value ofutility and discount the future at rate � 2 (0; 1).Political party j, if it wins power, has per-period utility function,

Uj = Wj �Mj; j = R;L

and �Mj otherwise, where W j is the gain in utility for party j if it wins theelection and Mj represents the amount of rents (income) transferred by party j

to other agents in the economy so that neither party is liquidity constrained. The

price that a party o�ers for the vote of an agent will in general depend on the

10We use a variant of a model which has become standard in the development literature, seefor example Banerjee, Gertler and Ghatak (2002)

8

occupation of the agent: let pj` be the price paid by party j to a landlord, pjw be

the price paid by party j for the vote of a worker, and pjs be the price paid for the

vote of a self-employed agent.11 Let � be the impact of one vote in party j's favor

on party j's chances of winning the election.12 From this we can deduce that the

maximal price that party j would be prepared to pay for a vote is �W j.

2.2. Timing of the stage game

The stage game has the following timing:

� The political parties non-cooperatively announce a price at which they willpurchase votes from each type of rural agent.

� The land market opens with each landlord deciding how much land to buy.

� Landowners hire workers by proposing a contract.

� Agents sell votes to the political parties.

� Workers vote and choose their e�ort level.

� Production take place and the output shock ~� is observed.

� Landlords and the political parties observe voting behavior and the state ofnature.

� Rents are distributed by the political parties, wages are paid and workersmay be �red and consumption takes place.

We now characterize the stationary subgame perfect equilibrium of this game.

11Though we focus our analysis on situations where political parties directly purchase votes,the model is consistent with other interpretations. For example, instead of buying votes, partiesmay o�er policies which favor landlords, or give landlords elected positions.12In a previous version of the paper we provided microfoundations for this assumption with

an explicit model of voting under proportional representation and legislative bargaining (seeAusten-Smith, 2000, and Baron and Deirmeier, 2001).

9

3. Electoral Corruption and Resource Allocation

For ease of exposition we �rst characterize the outcome of political competition

for votes. We do so by assuming, as will be the interesting case, that landlords

control and sell the votes of their workers. In the next section we analyze the

circumstances under which this will happen in equilibrium. To keep the discussion

focused we assume that all landowners are right-wing while all other agents are

left-wing. In addition we assume that the right-wing party values winning more

than the left-wing party. This will have the implication that the right-wing party

will be prepared to pay more for votes than the left-wing party.

The political parties engage in Bertrand competition. We �rst consider the

situation in which the right-wing party will always wish to outbid the left-wing

party for votes. This implies that �WR � �WL + �L and the following prices areo�ered by the parties in equilibrium,

Party R o�ers

8>:pR` = �W

L � �RpRw = �W

L

pRs = �WL + �L

and,

Party L o�ers pL` = pLw = p

Ls = �W

L

In this case, for any price that the left-wing party proposes for votes, the right-

wing party is always willing to outbid that o�er for the three categories of rural

agents. As a result, in equilibrium, the left-wing party announces the maximal

price it is ready to pay for one vote, �WL: Given this price, landlords will be

willing to sell their own votes to the right-wing party provided they can achieve

the same utility level that they could by selling their votes to the left-wing party.

This implies that the right-wing party must o�er them a price at least equal to

pR` = �WL��R: Landlords will also sell the votes of their workers if they are given

the same price that is o�ered by the left-wing party, which is then the price the

right-wing party announces. Lastly, for the self-employed agents, the right-wing

party must compensate them for not voting for their own preferred party, which

implies that he has to pay a price pRs = �WL + �L to those agents.

10

Given these prices, all rural agents sell their votes to the right-wing party,

with right-wing landlords stipulating that their left-wing workers vote right-wing

in their employment voting contracts.

In the case where �WL + �L � �WR > �WL; then

Party R o�ers

8>:pR` = �W

L � �R;pRw = �W

L;pRs = �W

R;

and,

Party L o�ers

8>:pL` = �W

L;pLw = �W

L;pLs = �W

R � �L:It is no longer optimal for the right-wing party to outbid the left-wing party for the

votes of the self-employed agents. Now, rather than buying the votes of all rural

agents, the right-wing party buys the votes of the landlords and their workers,

but the self-employed sell their votes to the left-wing party. Here, moving from

being self-employed to becoming a worker leads to a switch in voting behavior.

Under either scenario we have the following result.

Proposition 3.1. It is cheaper for the right-wing party to buy votes from a

landlord than to buy votes directly from the self-employed.

This result follows immediately from the fact that in equilibrium pRw < pRs .

This proposition has the implication that it will never be pro�table for a rural

agent to become a political entrepreneur, buying votes from individuals and then

selling them to parties. For the rest of the paper we focus on the situation where

�WR � �WL + �L, the analysis of the other parts of the parameter space followdirectly.

4. Employment and Power

We solve for the stationary subgame perfect equilibrium of this game which is

best from the point of view of landlords. In general a strategy for a landlord is a

11

contract o�er at date t which speci�es wages as a function of ~�, a voting decision

and the history of play up to t. For a worker a strategy determines an e�ort and

voting decision as a function of the history and the contract o�ered at t.

We start by describing the optimal labor-voting contract. As is standard, we

endow the landlord with all the bargaining power with respect to workers and he

can therefore make take-it-or-leave-it contract o�ers to his worker(s) specifying

his expected voting behavior and e�ort level. As there are two dimensions to the

worker's behavior, there are four possible wages, corresponding to whether output

is high or low, and whether the worker is observed voting for the speci�ed party

or not. We assume that liability is limited so that wages must be non-negative.

To ensure maximal incentives, a landlord will optimally propose a wage, w, and

continued employment if output is high and the worker is not observed voting for

the wrong party. If output is low or the worker votes for the left-wing party, the

landlord will pay zero and �re the worker. We assume that if a worker is ever �red

by a landlord he is never employed again by a landlord and is thus perpetually

self-employed.

We focus here on the situation under which a worker is required by his landlord

to vote for the right-wing party. Given his voting behavior, the worker will exert

the optimal amount of e�ort if the following incentive compatibility condition

is satis�ed. Let Vw(" = e) be the value to the worker if he exerts e�ort, while

Vw(" = 0) is the value if the worker shirks. The worker will exert e�ort if

Vw(" = e) � Vw(" = 0)

First consider the value from exerting e�ort which is,

Vw(" = e) = h (w + �Vw(" = e)) + (1� h)

0@��w + �WL + �L

�1� �

1A� e: (4.1)Here, with probability h the realization of output is high, in which case at date

t the worker receives the wage w and is not �red. In consequence the worker gets

the continuation value �Vw(" = e). With probability 1 � h, even though theworker exerted e�ort, output is low. In this case the worker gets no wage and

12

is �red at date t, never to be re-employed. In this case from date t + 1 on, the

worker is self-employed getting an income of w in each period and also being able

to freely sell his vote to whichever party he wishes. The utility from this latter

action is maxfpRs ; �L + pLs g, i.e. the self-employed agent can sell his vote to theright wing party and sacri�ce the utility bene�t of voting for his preferred party,

or he can sell his vote to the left and get the utility bene�t �L. Now in equilibrium

we showed that pRs � �L+ pLs � �WL+�L and this explains the formula in (4.1).We now consider the value of shirking, which is

Vw(" = 0) = l (w + �Vw(" = 0)) + (1� l)

0@��w + �WL + �L

�1� �

1A (4.2)The interpretation of (4.2) follows immediately from the discussion of (4.1), noting

that now, since the worker is shirking, he does not incur any e�ort cost and high

output arises with probability l. Hence, solving for the value functions, exerting

e�ort is optimal if

w � ��w + �WL + �L

�+(1� �l)

h � l e (4.3)

Next there is the participation constraint which shows that the worker prefers

accepting a contract to his outside option. This implies

Vw(" = e) �w + �WL + �L

1� � or w �w + �WL + �L + e

h: (4.4)

The two constraints are su�cient to de�ne a contract eliciting high e�ort and

voting behavior that will be accepted by a worker. Indeed, if this contract is

o�ered, the worker never chooses to vote against his landlord's wishes (whether

with high or low e�ort), as this implies with certainty his getting a zero wage

and being �red. This yields a utility to the worker that lies below his reservation

utility, so that, ex ante, the worker is better o� not accepting this contract. As

a result, accepting the contract but voting freely is never chosen in equilibrium.

Therefore, the only relevant constraints to the landlord's problem are (4.3) and

(4.4). Note also that, to be optimal from the landlord's point of view, the contract

must be such that one of these two constraints holds with equality.

13

We now explicitly de�ne the labour rents that are attributable to the existence

of a moral hazard problem in production. To do this, we consider the wage rate

that would be o�ered by the landlord in a contract that stipulates the e�ort

level, but not the voting behavior. In this contract, the e�ciency wage under the

incentive compatibility constraint would be

~w = �w +(1� �l)

h � l e: (4.5)

We consider the situation under which, with this contract, the participation con-

straint is satis�ed, that is:

~w � w + e

h

: (4.6)

Moreover, to simplify the exposition, we assume that it is optimal to propose this

contract which induces high e�ort by paying an e�ciency wage even if it concedes

rents to workers. We de�ne r as the per period excess in expected utility compared

to the reservation utility, that is the per period labour rent under the incentive

contract:

r = h ~w � e� w =

l

h � l e� w!(1� �h);

by using equation (4.5). We next de�ne R as the present discounted value of labour

rents:

R = r + �hr +��h

�2r + ::: =

r

1� �h =

l

h � l e� w

As can be inferred from condition (4.6), the participation constraint is satis�ed if

and only if R > 0 under the e�ciency wage.

We need one more assumption before stating our main result.

Assumption 1: �WL > h��L=(1� h�).

Assumption 1, which comes from �WL > h�(�WL+�L), says that the value

of a vote to the landlord must be large, relative to the ideological bias of the

worker.

Proposition 4.1. It is optimal for the landlord to also control the political be-

havior of his worker if and only if R > �L:

14

Proof. First note that if the wage had to be increased by the full disutility

of voting behavior being controlled, namely �WL + �L, then it could never be

pro�table for the landlord to o�er a contract which controlled voting. This is

because expected output would be the same but a vote can only be sold for �WL.

The wage that has to be paid to deter cheating when the landlord decides to

also control voting behavior must satisfy (4.3). Using (4.5), we can write:

w � ~w � �(�WL + �L) (4.7)

This wage has also to satisfy the participation constraint (4.4) and, using (4.6),

we obtain:

w � ~w � �WL + �L

h� R

h: (4.8)

Given that the contract must be optimal, one of those two conditions holds with

equality. First consider the case where (4.7) holds with equality. Paying this wage

is then pro�table for the landlord if the expected increase in the wage is less than

the bene�t from controlling a vote �WL, i.e.:

�WL � h (w � ~w) � h���WL + �L

�which holds by assumption. Consider the situation such that the participation con-

straint is binding, so that (4.8) holds with equality. Paying this wage is pro�table

to the landlord if :

�WL � h (w � ~w) � �WL + �L �R;

which holds if and only if R > �L.

Intuitively, the simple fact that the labor market is plagued by moral hazard

and that liability is limited implies that the landlord must concede rents to his

workers. This occurs even though he is in a position to make take-it or leave-it

o�ers to the latter. As long as the rents exceed the ideological bias, the increase in

wages landlords must concede to their workers in order to also control their voting

behaviour is lower than the price they receive for these votes from the parties.13

13The surplus thus given to the workers also yields a comparative advantage to the employerin other spheres, such as the credit market. This argument has been used in part of the literatureon interlinked contracts.

15

It is interesting to observe that if workers are unideological, so that �L = 0, as

long as the political parties attach a positive value to votes and there are rents,

Proposition 2 follows without any other assumptions. Note, however, in this case,

equilibrium the price paid by the right-wing party for the vote of any rural agent

is identical and therefore it is no longer true that it is strictly cheaper to buy votes

through landlords.

We now have to consider whether it is optimal for landlords to pay the e�-

ciency wage w� which satis�es both (4.3) and (4.4) with at least one equality. To

understand this we �rst consider the optimal demand for labor in a farm of size l

with n workers. Pro�ts are,

h�g

l

n

!n� hw�n+ �WLn (4.9)

The �rst term in (4.9) is expected revenues, the second the expected wage bill,

and the third the political rents that the landlord gets from selling the votes of

his n workers at the price �WL. The optimal demand for labor is determined by

the �rst-order condition,

h�

g

l

n

!� g0

l

n

!l

n

!� hw� + �WL = 0 (4.10)

The equation (4.10) implicitly de�nes the optimal demand for labor as a function

of parameters, which we write n(l; �WL; w�).

It is always pro�table for the landlord to pay this e�ciency wage contract if

�

h � l

��g

l

n(l; �WL; w�)

!+ �WL � hw� � w (4.11)

We therefore assume that the expected increase in pro�t from workers exerting

e�ort, evaluated at the e�ciency wage, plus the rents from selling their votes must

be greater than the expected increase in the wage bill.

The model has interesting implications for the price of land, denoted �: In the

model landlords hold land while workers have no access to capital markets and

cannot purchase land. Nevertheless, landlords could buy land from each other.

16

The equilibrium price of a plot of land must now adjust so that pro�ts are zero

or,

h�g

l

n(l; �WL; w�)

!� hw� + �WL

!n(l; �WL; w�)

l= � (4.12)

Equation (4.12) implies the following result.

Proposition 4.2. In equilibrium the price of land incorporates political rents.

Acquiring land is desirable not only for productive purposes, but also for the

political rents attached to the political control of the workforce employed on it.

Equilibrium prices on the land market reect this mechanism.

It follows from Proposition 4.2 that a political reform which stops landlords

controlling the political behavior of their workers, such as the introduction of an

e�ective secret ballot, removes the ability of landlords to sell their votes and has

interesting comparative static e�ects.

Proposition 4.3. The introduction of a secret ballot leads to a fall in the price

of land and the vote share of the right-wing party.

If all agents had access to capital markets then there would be no land con-

centration and all land would be farmed by smallholders with no workers getting

rents. The fact that, with perfect capital markets, smallholders are always willing

to outbid landowners for land follows from the fact that, through the participa-

tion constraint, the economic rents that landlords transfer to workers exceed the

political rents they receive from parties. Therefore, even though it is still true

that the ability of landlords to sell votes increases their demand for land, land is

still more valuable to smallholders.14

The interaction of the market failures is crucial. Without moral hazard there

are no rents and even with imperfect capital markets electoral corruption would

not a�ect the price of land, as workers would then have to be fully compensated

for the control of their votes. At the same time, with moral hazard but no capital

market imperfections there is no ine�ciency either.

14The ine�ciency here stemming from imperfections in the capital market is related to theresults of Banerjee and Newman (1993), Legros and Newman (1996), Mookherjee (1997) andBanerjee, Gertler and Ghatak (2002)

17

5. A Study of Land and Power: Chile

5.1. The political impact of the 1958 electoral reform: an overview

Like most Latin American countries, upon gaining independence from Spain, Chile

adopted republican institutions. These became institutionalized in the 19th cen-

tury and elections determined presidential succession without military or other

intervention. Universal literate male su�rage was introduced in 1874 but voting

was not secret. Interestingly, the 1874 su�rage extension in Chile was opposed by

some more progressive Chileans as they \fully realized that in a predominantly

rural society with traditional landlord-peasant ties, the Conservatives would over-

whelm their opponents at the polls." (J. Samuel Valenzuela, 1985, see also Bauer,

1995, p. 30). The nineteenth century democracy collapsed in 1924 and the fol-

lowing period saw �ve military coups before democracy was restored in 1932. The

intervening period was dominated by Colonel Carlos Ib�a~nez. After 1932 demo-

cratic stability was based on an explicit compromise between the growing power

of urban groups and the power of the traditional landed elites. 15

5.1.1. Mechanisms of Control of Rural Votes

\Throughout the history of the Republic, the political inuence of the rural sector

in Chile was disproportionately greater than its size relative to the urban sector.

Congressional representation was heavily weighted in favor of rural districts where

the peasantry historically formed a pliable and controllable mass base for conser-

vative and reactionary groups" (Hellinger, 1978, p. 272). Landlords systemati-

cally controlled rural voting until the late 1950s. There is a consensus amongst

historians, political scientists and sociologists about how this system functioned

(see Kaufman, 1972, Bauer, 1975, Loveman, 1976, Petras and Zeitlin, 1968, and

Scully, 1992, ch. 4): \There was an absolute control of peasants by their pa-

trones, and elections in rural communes depended on the political preferences of

the landowners" (Millar, 1981, p. 172). Large landlords usually registered all

15The Chilean pact is discussed in more detail in Valenzuela (1978), Collier and Collier (1991,pp. 565-73) and Scully (1992, pp. 108-109).

18

their employees, by teaching them how to sign their names (as literacy was a

condition for vote registration). The day of the election, the employer would go

vote with all their employees. \This type of control is pervasive ... The situation

was publicly accepted, and it was even used as an argument in electoral legal

complaints, particularly in order to show that any result against the preferences

of the latifundistas was fraudulent, or to justify an unanimous electoral result in

a rural locality" (Millar, 1981, p. 173). Part of the political pact which developed

after the 1930s also involved the banning of agricultural unions, a policy which

allowed severe labor repression to be carried on in the countryside, often backed

by the police (Bauer, 1995, p.32).

In line with our model, the control of rural votes by landlords was made

possible by the relatively good working conditions of the inquilinos compared

to the possible alternatives: \They were free ... but they had no defence in the

face of expulsion; indeed, the threat of being cast out into the subproletariat of

migratory workers was the most powerful weapon at the landowner's disposal.

Most inquilinos families undoubtedly judged their welfare on the estate superior

to life outside or in the nitrate �elds of the northern desert." (Bauer, 1995, p. 28).

The patron-client relationship was very developed (see in particular Bauer, 1995).

Thus, \anyone seen visiting the home of a resident laborer would be immediately

approached and questioned by the owner, who reserved the right to expel him

from the property" (Swift, 1971, p. 37).

5.1.2. The political equilibrium in the 1950s

By the 1950s the political landscape in Chile was dominated by several main par-

ties. There were the traditional nineteenth century parties, the Conservatives,

Liberals and Radicals. The Conservatives and Liberals were furthest to the right

and united in most things except in their attitudes to the Church (the Conserva-

tives were closely associated with the Catholic Church). The Radicals were more

towards the center politically and were strongly anti-clerical. Also in the center,

though small in the 1950s, were the Christian Democrats. To the left were the

Socialists and then the Communists (the latter were o�cially banned between

19

1948 and 1958 though they competed under di�erent names). The landed oli-

garchy provided the traditional constituency of the two right-wing parties, the

Conservative and the Liberal (see, e.g., Gil, 1966 and Sinding, 1972). The existing

party system was shocked however by the return from exile of the former dicta-

tor Carlos Ib�a~nez as a populist presidential candidate in 1952. Ib�a~nez formed a

very heterogenous coalition of mostly leftist groups and capitalized on the general

disillusionment with the traditional parties.

Chilean electoral institutions in this period were based upon the D'Hondt

system of proportional representation for all elections, under the 1925 Constitution

(for more details, see Gil, 1966, Chapter 5, and Morodo, 1968). The constituencies

broadly coincided with the boundaries of Chile's provinces. Deputies were elected

for four year terms, senators for eight with half of the Senate being replaced every

four years. Prior to the reforms of 1958, parties issued their own ballot papers

and a closed list system was used. Thus to vote for the Socialist party, a voter

had to request the Socialist ballot which made it relatively easy to determine

his voting behavior.16 Until 1951 only literate males over the age of twenty-one

could vote; of those eligible to vote, approximately 50 percent usually registered,

and the vast majority of those registered cast ballots. Women were allowed to

exercise the franchise in installments, �rst for municipal elections in 1935, then

for congressional contests in 1951, and �nally for presidential races in 1952.17

5.1.3. The introduction of the secret ballot in 1958

There were several important electoral reforms undertaken in Chile in the late

1950s and early 1960s. The most important was Law 12.889 promulgated on May

31st 1958, amending the basic electoral law of 1925 (see Castro, 1941, p. 35 and

Cruz Coke, 1984, pp. 27-29 for a discussion of this law) and its most important

aspect was the introduction of the c�edula �unica (the uni�ed ballot). After 1958,

the voter received a single, o�cial ballot, which contained all party slates for

16Loveman (2001, pp. 222-3) provides a detailed discussion of how party provision of ballotsbefore 1958 facilitated electoral corruption.17Note that registration and vote turn-out are very close in Chile, since once an individual

registers, voting is mandatory.

20

any single type of election in his district and an open list system was adopted so

that voters did not have to respect any o�cial ordering of candidates. Another

important law of 1958 banned electoral pacts between parties for deputies and

councilmen (a 1962 electoral law extended this prohibition to senatorial elections).

The introduction of the secret ballot had an immediate impact on the balance

of political power in Chile. Loveman (1976, p. 219) notes, \The introduction

of a public ballot meant that landowners could no longer e�ectively control the

votes of rural labor. The electoral hegemony of the Right in the countryside thus

gave way to forces that advocated social change in the rural areas ... In 1958 the

performance of the FRAP (Socialists and Communists) in rural districts left little

doubt that landowners' control over rural votes had considerably declined."

If the lack of secret balloting had played an important role in guaranteeing

democratic stability in Chile since the 1930s, why was the secret ballot introduced

in 1958? Though this issue appears not to have been researched by political

scientists, the most plausible reason for this is a deliberate attempt to disrupt the

existing political equilibrium. As we noted above, the election of Ib�a~nez in 1952

was based on a heterogenous coalition and an `anti-politics' platform.18 Ib�a~nez

intended to forge a new political movement and though he failed in this, it seems

likely that the introduction of the secret ballot, with its easily anticipated e�ects

on voting in the countryside, was a calculated gamble. It may also have been part

of a deal which he made with some of his key supporters, the Agrarian Labor party

(Agrario Laboristas) and the Popular Socialist party (Partido Socialista Popular)

both of which would have had an interest in mobilizing rural voters.

Interestingly however, despite these changes, the Conservative Jorge Alessan-

dri won the presidential election in 1958, principally on a platform emphasiz-

ing conservative monetary policies which were a response to the populism of the

Ib�a~nez regime.19 Under Ib�a~nez per-capita GDP had fallen by 2% and ination

had averaged 45%, peaking at an annual rate of 76% in 1955 (see Ffrench-Davis,

18His campaign was based on the symbol of a "broom" with which he promised to "sweep"away political corruption and bad governments...19Though he polled only 33,416 votes (out of 1,235,552 cast) more than Salvador Allende,

the candidate for the Socialist and Communist alliance. Antonio Zamorano, a leftist defrockedpriest, deprived Allende of a victory by gaining 41,304 (3.3%) left-wing votes.

21

1973, p. 242 and Table 35). However, the right began to disintegrate during the

1960s with the rise of the centrist Christian Democratic Party (Partido Dem�ocrata

Cristiano{PDC), founded in 1957 with the merger of three conservative elements:

the National Falange, founded in 1938; the Social Christian Conservative Party;

and the remnants of the Agrarian Labor Party that had backed Ib�a~nez. In the

1960s, the Christian Democrats became more progressive, and espoused reformist

Catholic doctrines, which appealed strongly to the middle class, women and rural

voters. In 1966 the Conservatives and Liberals merged to form the National Party.

The Christian Democrat candidate, Eduardo Frei, won the presidency in 1964, and

in 1970 the Christian Democrats provided support to Allende, in exchange of his

guaranteeing strict adherence to democratic procedures.

5.2. Agrarian relations and electoral results across provinces

We collected data from the 1957, 1961 and 1965 electoral registries, the agri-

cultural census of 1935, 1955 and 1965 (see the Appendix) and the population

censuses of 1930 and 1940. The data were collected at the communa (municipal-

ity) level, which corresponds to the lowest level of electoral district as well as the

smallest administrative unit. As census units do not always strictly match the

electoral districts, and changed de�nition over time, we had to exclude all the

communas for which we could not be certain of the correspondence, which left us

with a sample of 246 communas (out of 295). (This also explains why we do not

use data over a longer interval of time.) The main variables used throughout the

analysis are described in the Appendix (table A1).

We focus on the 1957 parliamentary (all of the congress and half of the senate)

elections, as it allows a more direct comparison to the parliamentary elections that

occurred after 1958, and in particular the 1961 and 1965 elections.20 We describe

20We do not take the 1953 congressional and senatorial elections because they were veryexceptional. After the election to the presidency of Carlos Ib�a~nez in 1952, the 1953 electionssaw a transient collapse in the right-wing vote in the face of the Ib�a~nista bandwagon. Scully(1992, p. 126) notes \The disruption of familiar patterns of party competition was also reectedin the extreme fragmentation by the party system in the congressional elections of 1953. Inthat year, 25 party organizations presented candidates, and 19 achieved representation. Partyproliferation weakened Chile's traditional parties. Whereas in the congressional elections of 1949

22

in Table 1 the main trends at the level of the provinces. In the table, we report

the information over the three Central Valley regions, its two neighboring regions,

the Frontier and the Little North, and the other regions.

INSERT TABLE 1 HERE

The relationship between agrarian relations and electoral outcomes is striking.

In 1957, the landed oligarchy in Chile dominated the Urban and the North Central

Valley provinces: the proportion of inquilinos in the number of registered voters

in 1957 is 18.9 percent in the North Central Valley, and 17.2 in the Urban Central

Valley, but 11.2 in the Frontier and Little North, and 8.2 in the other provinces.

Unsurprisingly, the share of right-wing votes in 1957 in the North Central Valley

was 50.0 percent, and 40.8 percent in the Urban Central Valley, much higher than

in the other provinces.21

After 1958, the fall in the right-wing votes occurs in provinces with a larger

proportion of inquilinos per worker (across provinces, the correlation coe�cient

between the two is equal to -0.67). The fall in right-wing votes is dramatic in

the Central Valley provinces. Even the absolute number of right-wing votes fell in

those areas, in spite of an increase in registered voters. The fall is very pronounced

in some provinces, such as Colchagua (-48.1%) from an absolute majority of 70.2%

of the votes in 1957 to barely 22.5 % in 1965.22

the Conservative, Liberal, and Radical parties combined received more than 60% of the vote,in 1953 they received barely on third." This was just a temporary phenomena however. Scullygoes on to add (1992, p. 126) \Though Ib�a~nez had put the leadership of traditional parties onthe defensive in 1953, the situation was reversed between 1953 and 1957." Focusing on 1953therefore has a tendency to underestimate the fall in conservative support after 1958. Moreoverthe de�nition of electoral circonscriptions changed through time, which would further restrictour sample.21The relationship between right-wing votes in the 1957 elections and land concentration is

less clear however. This is due to the fact that in the arid, semi-arid and infertile provinces tothe north and to the south of the Central Valley (including the Frontier), land concentrationtends also to be high, as a result of the technological constraints on agriculture in these provinces(ranching instead of farming).22Hellinger (1978) analyzed electoral change in the Chilean countryside for the two presidential

elections of 1958 and 1970, based on correlations from a restricted (and biased) sample ofmunicipalities. He points out the gradual erosion of support for the Right as the correlationbetween the vote for the Nationalists as the proportion of inquilinos in the agricultural workforcefell. He however fails to provide a consistent explanation for this change in rural voting pattern.

23

5.3. The political impact of the 1958 electoral reform: a test

5.3.1. The empirical strategy

The empirical strategy pursued in this paper can be described as follows. Before

the 1958 reform, the share of right-wing votes should be higher in communas

with more inquilinos since their votes are then controlled. However, after the

reform, the inuence of inquilinos on electoral results should disappear, so that the

di�erence in voting patterns across the two types of communas should disappear.

In table 2 below, we report the electoral results in 1957 and 1965 for communas

with less and more inquilinos than the median.

INSERT TABLE 2 HERE

Over the period, right-wing votes in communas with less inquilinos fell by -

16.2 % while it fell by -30.3% in communas with more inquilinos. The impact of

the loss of control over inquilinos votes on the fall in right-wing votes corresponds

to the di�erence between these two �gures, -14.1%. The model below aims at

estimating this impact more precisely.

In Figure 1, we present simple OLS scatter plot of the relationship between

right-wing votes and the proportion of inquilinos in each communa. The pattern is

striking, as the impact of inquilinos on right-wing votes is signi�cantly diminished

after 1958.

INSERT FIGURE 1 HERE

5.3.2. The empirical models

We now present the model used in the empirical analysis. We let RVi;t represent

the number of votes cast in favor of the right-wing party, Vi;t, the total number

of voters, and V hi;t, the total number of voters of type h at time t in communa i.

Voters can be of three di�erent types: h = inq if the voter is an inquilino, h = agr

if the voter is not an inquilino but works in agriculture, and h = na if he is not

an agricultural worker. We can then write:

RVi;t = (�i+�I;t+�inqt )V

inqi;t +(�i+�I;t+�

agrt )V

agri;t +(�i+�I;t+�

nat )V

nai;t +"i;tVi;t (5.1)

24

where �i is a communa speci�c �xed e�ect, which represents the time-invariant

propensity to vote for the right-wing party in that communa, �I;t is a provincial

level �xed e�ect at each time period which represents the propensity to vote

for the right-wing party in province I at time t; and �ht represents the speci�c

propensity for a voter of type h to vote for the right-wing party at time t. The

error component, "i;t, satis�es the usual conditions. Rearranging equation (5.1)

above, and using the fact that Vi;t = Vinqi;t + V

agri;t + V

nai;t ; we obtain:

RVi;t = (�I;t + �nat )Vi;t + (�

inqt � �nat )V

inqi;t + (�

agrt � �nat )V

agri;t + (�i + "i;t)Vi;t:

Dividing both sides of the equation by Vi;t, one gets:

RVi;tVi;t

= �I;t + �nat + (�

inqt � �nat )

V inqi;tVi;t

+ (�agrt � �nat )V agri;tVi;t

+ �i + "i;t; (5.2)

that potentially can be directly estimated. However, we have no information on

the number of voters per category of occupation, V hi;t. Moreover, even between

1957 and 1965, the number of registered voters in the population varied.

To address these issues we have to make assumptions in order to determine the

proportion of voters per occupational category, and we will investigate di�erent

alternatives. In Model 1, we assume that the change in registration is homogenous

across all social classes in a given communa, but may di�er across communas: Vi;tVi;t0

is speci�c to communa i. Moreover, we also assume that, at time t0, the probability

that an inquilino is registered as an elector is the same across all communas.23

Letting N inqi;t represent the number of inquilinos in communa i at time t, we can

then write:V inqi;t

N inqi;t=Vi;tVi;t0

V inqt0N inqt0

,V inqi;tVi;t

=V inqt0N inqt0

N inqi;tVi;t0

:

23As we shall see, this assumption is necessary to identify the impact of the prevalence ofinquilinos in the electorate. If the proportion of inquilinos voting in 1957 is arbitrary, identi�ca-tion becomes impossible, as the impact of the electoral reform on the 1965 electoral results canalways be explained by di�erences in the registration of inquilinos in 1957 between communas.

25

whereV inqt0N inqt0

is not speci�c to communa i, and Vi;tVi;t0

, the change in registration rate

in communa i, is not speci�c to an occupational category.

We make a similar assumption for the other agricultural workers (though our

main tests do not require this), where Nagri;t represents the number of agricultural

workers other than inquilinos in communa i at time t:

V agri;tVi;t

=V agrt0Nagrt0

Nagri;tVi;t0

:

Using these two expressions in equation (5.2), and rearranging those terms, we

obtain:

RVi;tVi;t

= �I;t+�nat +

(� inqt � �nat )

V inqt0N inqt0

!N inqi;tVi;t0

+

(�agrt � �nat )

V agrt0Nagrt0

!Nagri;tVi;t0

+�i+"i;t;

(5.3)

which represents the basic equation to be estimated. For further interpretation, it

is convenient to rewrite the latter by considering only two time periods, 1957 and

1965, as well as by introducing explicitly provincial dummies, DI , (which is equal

to 1 if communa i belongs to province I and zero otherwise) and a time dummy

t65 corresponding to year 1965. We then have:

RVi;tVi;t

= �i +XI

�I;57DI + �na57 +

XI

�I;�65DIt65 + �na�65t65

+�57N inqi;tVi;57

+ ��65N inqi;tVi;57

t65 + 57Nagri;tVi;57

+ �65Nagri;tVi;57

t65 + "i;t; (5.4)

where �i +PI �I;55DI + �

na57 ; represents the (total) communa level �xed e�ect.

The basic test we want to carry out is the following. Before the secret ballot was

introduced, �57 > 0 and �57 > 57 so that the right-wing party gets more votes

in a communa with more inquilinos compared to other voters, and particularly

other agricultural workers. The main impact of the electoral reform is that, after

1958, inquilinos can vote freely. We therefore expect ��65 < 0; as the inuence of

inquilinos on the electoral results of the right-wing party should fall signi�cantly

after 1958. We also expect that inquilinos in 1965 will vote like the other agri-

cultural workers, so that �57 + ��65 = 57 + �65 (or = 0 if we assume that the

26

other agricultural workers do not vote di�erently than the non-agricultural social

classes): This last expression represents our main empirical test.

As said above, the number of registered voters increased during those years.

In Model 1, we assume that, within a commune, the increase in registration is

identical across all social classes. However, if non-inquilinos are more likely to

be registered in 1965 than in 1957, this can then potentially introduce a bias in

our estimates if the registration of non-inquilinos is negatively correlated with the

proportion of inquilinos in the voting population at time t0. In this case the 1965

e�ect we are capturing may simply be the e�ect of an increase in registration that

is biased against inquilinos. This is a di�cult issue, to which we cannot provide

a simple answer. (Note however that what this argument points out is that the

e�ect of the electoral reforms on the 1965 elections may be over-estimated. But it

leaves unbiased the coe�cients associated with the inquilinos before the reform.)

Model 2 assumes that, across all communas and across time, the proportion

of voting inquilinos in the inquilino population remains constant:

V inqi;t

N inqi;t=V inqt0N inqt0

,V inqi;tVi;t

=V inqt0N inqt0

N inqi;tVi;t

This assumption implies that any increase in the proportion of registered voters

takes place among non-inquilinos (an extreme version of the above argument).

Under this assumption, it seems more reasonable to neglect the potential di�er-

ence in voting behavior between the two other classes, the explicit incorporation

of which requires additional assumptions to determine which classes bene�tted

most from the increase in registration. (In e�ect, we therefore assume that the

other agricultural workers do not vote di�erently from the other non-agricultural

classes.) Note that we still assume that, at time t0, the probability that an in-

quilino is registered as an elector,�V inqt0N inqt0

�; is not communa-speci�c. Rewriting

equation (5.3) under these assumptions, we obtain:

RVi;tVi;t

= �I;t + �ninqt +

(� inqt � �ninqt )

V inqt0N inqt0

!N inqi;tVi;t

+ �i + "i;t; (5.5)

27

or, following eq(5.4):

RVi;tVi;t

= �i +XI

�I;57DI + �ninq57 +

XI

�I;�65DIt65 + �ninq�65 t65

+�57N inqi;tVi;t

+ ��65N inqi;tVi;t

t65 + "i;t: (5.6)

The equation above can also be directly estimated, since the RHS variable is

simply the number of inquilinos at time t divided by the number of voters at time

t (instead of at time t0 as in model (1)). Again, the strategy consists in testing

�57 > 0; ��65 < 0; and �57 + ��65 = 0:

In Model 3, we assume that the number of voting inquilinos remains constant

throughout the period, while keeping the assumption that, at time t0, the prob-

ability that an inquilino is registered as an elector is not communa-speci�c. We

have:V inqi;tVi;t

=V inqi;t0Vi;t

=V inqt0N inqt0

N inqi;t0Vi;t

:

This implies that any increase in the number of registered voters took place among

non-inquilinos. Once again neglecting potential voting di�erences between the

other agricultural workers and the non-agricultural classes, equation (5.3) be-

comes:

RVi;tVi;t

= �I;t + �ninqt +

(� inqt � �ninqt )

V inqt0N inqt0

!N inqi;t0Vi;t

+ �i + "i;t; (5.7)

which we rewrite:

RVi;tVi;t

= �i +XI

�I;57DI + �ninq57 +

XI

�I;�65DIt65 + �ninq�65 t65

+�57N inqi;57Vi;t

+ ��65N inqi;57Vi;t

t65 + "i;t: (5.8)

Equation (5.8) can also be directly estimated, using the number of inquilinos in

communa i in 1957 divided by the total number of voters in that communa at

time t. An extreme version of this model requires that, in 1957, all inquilinos were

28

registered, thereby implicitly assuming thatV inqt0N inqt0

= 1 in equation (5.7) above, but

leads to the same speci�cation as in equation (5.8).

In Model 4, we assume that the probability that an agent votes in municipality

i is identical across all classes, and equal to the share of agents of that class in

the population of the municipality (proportional registration). Registration rates

are communa speci�c, however. We then have:

V inqi;tVi;t

=N inqi;tNi;t

;V agri;tVi;t

=Nagri;tNi;t

;

where Ni;t represents the total population in that communa at time t. Combining

the latter expressions with equation (5.3), one gets:

RVi;tVi;t

= �I;t + �nat +

�� inqt � �nat

� N inqi;tNi;t

+ (�agrt � �nat )Nagri;tNi;t

+ �i + "i;t; (5.9)

or; more explicitly:

RVi;tVi;t

= �i +XI

�I;57DI + �na57 +

XI

�I;�65DIt65 + �na�65t65

+�57N inqi;tNi;t

+ ��65N inqi;tNi;t

t65 + 57Nagri;tNi;t

+ �65Nagri;tNi;t

t65 + "i;t: (5.10)

This equation requires to use the proportion of inquilinos in the population of

the municipality at time t as the measure of voting inquilinos. Note here that

assuming proportional registration for the other agricultural workers allows us to

introduce them separately in our estimation, as in equation (5.4).

Model 5 requires that the proportion of voting members of a particular class

in the voting population is constant across time. As in Models 1, 2 and 3, we still

assume that the probability that a member of a particular class is a voter at time

t0 is not speci�c to a particular municipality. We have:

V inqi;tVi;t

=V inqi;t0Vi;t0

=V inqt0N inqt0

N inqi;t0Vi;t0

;V agri;tVi;t

=V agri;t0Vi;t0

=V agrt0Nagrt0

Nagri;t0Vi;t0

:

29

Note that this assumption still allows inquilinos to be over-represented in 1957.

A variant of Model 4 is to assume that, in 1957, inquilinos are over-represented

as voters to the point that all inquilinos are registered as voters, while, after the

1958 reform, the probability of voting is identical across classes. To simplify, we

also ignore di�erences in voting behavior among the remaining social classes. We

then have:

V inqi;tVi;t

=N inq57Vi;57

in 1957 andV inqi;tVi;t

=N inqi;tNi;t

in 1961, 1965:

This speci�cation will be referred to as Model 6 in the estimations below.

It is clear that not all models are equally defensible. Model 6 allows for the pos-

sibility of biased registration by landlords, whereby inquilinos are over-represented

in the voting population in 1957. This is also a feature of Models 1, 2, 3 and 5.

However, these last three models assume some stability for the subsequent elec-

tions: Model 1 assumes that the proportion of voting inquilinos increased at the

same rate as for the other social classes, Model 2, that the proportion of voting

inquilinos in the inquilino population remained constant, Model 3, that the num-

ber of voting inquilinos remained constant, and model 5 that the proportion of

voting inquilinos in the voting population remained constant. By contrast, Model

6 implies that a large number of inquilinos lost their registration as voters, which

runs against the facts. Indeed, even in 1965, voters represented only 25% of the

total population: this actually implies that, between 1957 and 1965, the number

of inquilinos increased four times or that more than 3/4 of the inquilinos were

de-registered ! (The number of inquilinos actually declined by 8% over the pe-

riod.) Similarly, the assumption in Model 3 that the number of voting inquilinos

remained constant is also not very satisfactory. Model 4 does not allow for biased

registration in favour of inquilinos before 1958. And Model 5 goes against the

idea that increased registration after 1958 mostly favored the other social classes.

As a result, the most relevant models are Models 1 and 2. As Model 1 allow us

to explicitly incorporate the other agricultural workers without any additional

assumptions, we will refer to it as our basic model. The other models will be

used mainly to show that our main results are not too sensitive to the particular

assumptions made on the registration process of inquilinos.

30

5.3.3. The basic results

The information we have on inquilinos, total population, and the agricultural

labour force in a municipality comes from the two agricultural censuses of 1955

and 1965. For the elections of 1965, we used the information from the 1965 census,

while for the elections of 1957, we computed the corresponding �gures, using linear

interpolation between 1955 and 1965.

The results of the panel estimations, with a communa �xed e�ect, are given

in Table 3 below. Column (1) of Model 1 corresponds to equation (5.4) above,

without the province*year dummies. By doing so, we impose homogeneity on the

evolution of the votes across provinces (electoral results vary in the same way

in all provinces and are therefore picked by the year dummy), thereby assuming

that: �I;�61 = �I;�65 = 0; 8I:In column (2), we allow provincial e�ects to vary with time, as a robust-

ness check, as well as to reect possibly changing state policies that could a�ect

provinces di�erently (for instance, between cattle-raising and grain-growing ar-

eas). The inclusion of time-varying province e�ects may however a�ect the sig-

ni�cance of our estimates, because of the provincial pattern in the prevalence of

the inquilino system (i.e. multicollinearity).24 In column (2), we also include the

following additional controls: total population and the proportion of land under

large farms in the municipality. Column (3) presents the same speci�cation as in

column (1), but uses pooled OLS with a province �xed e�ect.

The results obtained using Model 2 is given in columns (4) and (5), with

similar speci�cations as in columns (1) and (2). The corresponding estimates under

Models 3, 4, 5 and 6 are given in columns (6)-(14).

INSERT TABLE 3 HERE

The results are striking, as they tend to con�rm all the hypotheses made above.

First, communas with more inquilinos are more likely to display stronger support

in favour of the right-wing party in 1957. Thus, following column (1), in 1957,

in a communa where the share of inquilinos in the number of voters is greater by

24As can be expected, the estimation results tend to be stronger in the absence of communa�xed e�ects (using pooled OLS) or in the absence of province*time dummies.

31

one standard deviation (0.14), the expected share of the right-wing party is higher

by 6.1%. This represents a 15% increase in the average proportion of right-wing

votes. It is worth noting that the coe�cient attached to inquilinos in 1957 di�ers

signi�cantly from that attached to the other agricultural workers, which implies

completely distinct voting behaviors between the two classes (and in many cases,

�57 > 0 and 57 < 0) .

Moreover, the political inuence of inquilinos disappears in 1965. The coe�-

cient estimated, ��65; is negative and signi�cant, but �57 + ��65 does not di�er

from 57 + �65; the corresponding coe�cients for the other agricultural work-

ers, nor from 0 which corresponds to the voting behavior of the non-agricultural

classes. The imposition of secret ballot therefore had an important and signi�cant

e�ect, as the impact of inquilinos on right-wing votes completely vanishes in the

1965 elections. 25

The results are consistent over the alternative speci�cations. The main coe�-

cients of interest always have the anticipated sign, but sometimes lose signi�cance.

In particular, the results are weaker when we control for the interaction of province

and time e�ects. This is a very demanding requirement however since in such a

model we are only using the intra-province variation to identify the causal e�ect

and this speci�cation also introduces a problem of multicollinearity with the vari-

ables of interest. Models 3 and 6 also produce less satisfactory results, however

these are the two models based on the least plausible identi�cation assumptions.

To further test the robustness of the above results, we ran similar regressions

using other indicators of the strength of patron-client relationships and of political

control by a traditional landed oligarchy. Instead of using the proportion of voters

of di�erent types in the voting population, we used the proportion of inquilinos in

the agricultural labour force in 1957 and 1965 as a measure of the intensity of the

patron-client relationships in the communa, and a measure of land concentration,

the share of area owned by farms larger than 200 hectares in the total agricultural

25Although not reported here, the provincial dummies attached to the `oligarchic' provincesof O'Higgins, Aconcagua and Colchagua are always signi�cant for the 1965 elections (equal to-0.16, -0.19 and -0.26 respectively, all signi�cant at the 1% level). (While we cannot estimate theprovincial �xed e�ect for 1957 with the panel regressions, the corresponding estimates obtainedwith the pooled OLS for 1957 are 0.15, 0.27 and 0.33, all signi�cant at the 1% level).

32

area of the communa.26 We report the results of these estimations in table 4

below. The estimates, though less precise than in the basic model, are once again

entirely consistent with our main hypotheses. It must be noted, however, that

many coe�cients associated with land concentration or agrarian relations in 1965

loose their signi�cance when interacted time and provincial dummies are included

in the model, though they remain consistently negative. This once again may be

partly attributed to the multicollinearity with the interacted time and provincial

dummies, as well as to the less precise nature of the indicators used. Interestingly,

when we run a regression using both the proportion of inquilinos and the measure

of land concentration as in column (17), the latter looses all signi�cance, contrary

to the former. This suggests that land concentration had less implications for

the political outcome of a municipality than the intensity of the patron-client

relationship, which is perfectly consistent with our model.27

The last three columns of table 4 present the results obtained when one uses

the proportion of inquilinos in the population of a municipality in 1935. One can

argue that this variable is a good measure of the importance of inquilinos, as it

predates the elections under study, and is therefore less susceptible of a possible

endogeneity bias. The results fully support our assumptions. Note that, with a

panel �xed e�ect, we cannot estimate the coe�cients attached to variables which

remain constant over time, in particular the one related to the 1957 elections.

They are estimated using pooled OLS in column (20).

INSERT TABLE 4 HERE

The above estimates excluded the 1961 elections. The problem comes from

the fact that the number of inquilinos are observed only in two years, 1955 and

1965, so that our main estimations cover the two closest election dates, 1957

and 1965. We now use the 1961 electoral results, where the number of inquilinos

26Again the 1957 �gures were obtained by linear interpolation between 1955 and 1965. Theseland concentration measures are imprecise however, as the censuses only report at the communalevel the number of farms per size category. By taking the median of each size class, we computedan estimate of the areas, that we used to compute the shares in area.27Note that large farms were also found in cattle-raising areas in the North and in the South

of Chile, where few inquilinos were found. The correlation between land concentration and theproportion of inquilinos in the agricultural labour force is only 0.22.

33

and the other relevant variables for 1961 were obtained by linear interpolation

between 1955 and 1965. In table 5, we present the results obtained under the �ve

di�erent models discussed above. Note that, since the 1961 elections were close to

the electoral reform and all political adjustments might not be instantaneous, we

expect 0 > ��61 � ��65: we expect the impact of the electoral reform to developmore fully in 1965 than in 1961. The results are presented in Table 5 below.

INSERT TABLE 5 HERE

The results follow closely our former results. The coe�cients, and their stan-

dard errors, associated with the 1957 and 1965 elections are almost identical. For

the 1961 elections, the results are less conclusive, as the coe�cient ��61 associated

with the inquilinos in 1961 is almost everywhere negative, but not systematically

signi�cant. As expected, the coe�cients associated with the 1961 election are

smaller (in absolute value) than the corresponding ones in 1965. The last column

presents the results obtained when using the proportion of inquilinos in the pop-

ulation in 1935, with signi�cant and negative coe�cients for the 1961 and 1965

elections.

5.3.4. Christian Democratic votes

We ran similar regressions using the vote share of the Christian Democratic party

as the dependent variable. One can argue that, after the introduction of the secret

ballot, inquilinos are more likely to vote for that party than for any other party.

(We also ran similar regressions on the socialist and the Christian Democratic

parties together, with very similar results.) The focus on the Christian Democratic

party follows from the fact that, historically, peasant and female voters constituted

their political base, while the socialist party drew more support from men and

blue collar workers.

The estimates are given in table 6 below, where we followed the speci�cation

used in the six models above. The results are once again supportive of our hypothe-

sis. While, before 1958, the communas with a higher proportion of inquilinos tend

to vote less in favour of the left-wing parties, this impact completely disappears

in 1965. The estimates are consistent across the various speci�cations, though

34

not always signi�cant, particularly when time*province dummies and additional

controls, such as land concentration or population are used.

INSERT TABLE 6 HERE

6. Conclusions

In this paper we have investigated how the employment relationship, if it gener-

ates rents, may allow employers to control the political behavior of their workers.

The salient example of this is voting behavior when there is no secret ballot, so

that political behavior is observable. Interestingly, the rents conceded by em-

ployers to workers gives the former a comparative advantage in controlling the

political activities of the latter, relative to political parties. We showed that this

helps to explain one of the big stylized facts about polities with endemic electoral

corruption, namely that employers supply votes to parties rather than the parties

buying most votes separately from individuals. The ability to sell votes increases

the demand for labor and generates an added incentive to own land, driving up

its price.

We test the predictions of the model by examining in detail the e�ects of the

introduction of the secret ballot in Chile in 1958. We show that, consistent with

our theory, the political reforms led to large changes in voting behavior. Before the

reforms, localities with more pervasive patron-client relationships tend to exhibit

a much stronger support for the right-wing parties, traditionally associated with

the landed oligarchy. After the reform however, this di�erence across localities

completely disappeared. This evidence suggests to us that electoral corruption,

and the economic and political incentives that it created, is an important part of

the story for why inequality has been so high historically in Latin America and

possibly also an important part of the story about why long-run economic perfor-

mance in Latin American has been so disappointing (on which see Engerman and

Sokolo�, 2005, and Posada-Carb�o, 2000, who argue for the central importance

of electoral corruption in Latin American political history). Though our analysis

focused on vote buying, this can be thought of as a metaphor for a wide variety

of political favors or policies that transfer rents to the landlords. Moreover, the

35

political control that rents allow employers to exercise applies much more gen-

erally, even in situations where there is an e�ective secret ballot. Any type of

observable political activity, collective actions, demonstrations, trade unionism,

political activism, can all be controlled by the threat of losing ones employment

and the rent that it provides.

36

References

Acemoglu, Daron and James A. Robinson (2006) Economic Origins of

Dictatorship and Democracy, New York; Cambridge University Press.

Anderson, Margaret L. (1993) \Voter, Junker, Landrat, Priest: The Old

Authorities and the New Franchise in Imperial Germany," American Historical

Review, 98, 1448-1474.

Austen-Smith, David (2000) \Income Redistribution under Proportional

Representation," Journal of Political Economy, 108, 1235-1269.

Banerjee, Abhijit V., Paul J. Gertler and Maitreesh Ghatak (2002)

\Empowerment and E�ciency: Analysis of a Tenancy Reform in West Bengal,"

Journal of Political Economy, 110, 239-280.

Banerjee, Abhijit V. and Andrew F. Newman (1993) \Occupational

Choice and the Process of Development," Journal of Political Economy, 101, 274-

298.

Baron, David P. and Daniel Diermeier (2001) \Elections, Governments,

and Parliaments in Proportional Representation Systems," Quarterly Journal of

Economics, 933-967.

Bauer, Arnold J. (1975) Chilean Rural Society from the Spanish Conquest

to 1930, Cambridge University Press, Cambridge UK.

Bauer, Arnold J. (1995) \Landlord and Campesino in the Chilean Road

to Democracy," in Evelyn Huber and Frank Sa�ord eds. Agrarian Structure and

Political Power: Landlord and Peasant in the Making of Latin America, University

of Pittsburgh Press, Pittsburgh, PA.

Bendix, Reinhard (1964) Nation-Building and Citizenship, John Wiley and

Sons, New York.

Blackbourn, David (1998) The Long Nineteenth Century: A History of

Germany, 1780-1918, Oxford University Press, New York NY.

Breman, Jan (1974) Patronage and Exploitation: Changing Agrarian Rela-

tions in South Gujarat, India, University of California Press, Berkeley CA.

Buchanan, James and Gordon Tullock (1962) The Calculus of Consent,

Ann Arbor; University of Michigan Press.

37

Castro, Jos�e Luis (1941) El Sistema Electoral Chileno, Nascimento, Santi-

ago.

Collier, Ruth Berins and David Collier (1991) Shaping the Political

Arena, Princeton University Press, Princeton NJ.

Cox, Gary W. (1987) The E�cient Secret: The Cabinet and the Develop-

ment of Political Parties in Victorian England, Cambridge University Press, New

York NY.

Cox, Gary W. (1997) Making Votes Count, Cambridge University Press,

New York, NY.

Cruz-Coke, Ricardo (1984) Historia Electoral de Chile, 1925-1973, Edito-

rial Juridicia de Chile, Santiago.

Dekel, Eddie, Matthew O. Jackson and Asher Wolinsky (2004) \Vote

Buying," Unublished California Institute of Technology.

Engerman Stanley L. and Kenneth L. Sokolo� (2005) \The Evolution

of Su�rage Institutions in the New World," Journal of Economic History, 65,

891-921

Ffrench-Davies, Ricardo (1973) Pol��ticas Econ�omicas en Chile, 1952-

1970, Ediciones Nueva Universidad, Santiago Chile.

Gash, Norman (1977) Politics in the Age of Peel, Harvester Press, Hassocks

UK.

Gibson, Ralph and Martin Blinkhorn (1991) Landownership and Power

in Modern Europe, Harper Collins, New York NY.

Gil, Frederico G. (1966) The Political System of Chile, Houghton Mi�in,

Boston MA.

Goldstein, Richard J. (1983) Political Repression in Nineteenth Century

Europe, Croon Helm, London UK.

Graham, Richard (1990) Patronage and Politics in Nineteenth Century

Brazil, Stanford University Press, Stanford CA.

Grossman, Gene M. and Elhanan Helpman (1996) \Electoral Compe-

tition and Special Interest Politics," Review of Economic Studies, 63, 265-286.

Hamerow, Theodore S. (1974) The Social Foundations of German Uni-

�cation, 1858-1871: Struggles and Accomplishments, Princeton University Press,

38

Princeton NJ.

Hartlyn, Jonathan and Arturo Valenzuela (1998) \Democracy in Latin

America since 1930," in Leslie Bethell ed. Latin America: Politics and Society

since 1930, Cambridge University Press, Cambridge UK.

Hellinger, Daniel (1978) \Electoral Change in the Chilean Countryside:

the Presidential Elections of 1958 and 1970", The Western Political Quarterly,

31(2), 253-73.

Kaufman, Robert E. (1972) The Politics of Land Reform in Chile, 1950-

1970: Public Policy, Political Institutions, and Social Change, Harvard University

Press, Cambridge MA.

Kinzer, Brian L. (1982) The Ballot Question in Nineteenth Century English

Politics, Garland, New York NY.

Kitson-Clark, G.S.R. (1951) \The Electorate and the Repeal of the Corn

Laws," Transactions of the Royal Historical Society, 1, 109-126.

Kohli, Atul (1990) Democracy and Discontent: India's Growing Crisis of

Governability, Princeton University Press, Princeton NJ.

Kreuzer, Marcus (1996) \Democratization and Changing Methods of Elec-

toral Corruption in France from 1815 to 1914," in Walter Little and Eduardo

Posada-Carb�o eds. Political Corruption in Europe and Latin America, MacMil-

lan, London.

Legros, Patrick and Andrew F. Newman (1996) \Wealth E�ects, Dis-

tribution, and the Theory of Organization," Journal of Economic Theory, 70,

312-341.

Loveman, Brian (1976) Struggle in the Countryside: Politics and Rural

Labor in Chile, 1919-1973, University of Indiana Press, Bloomington IN.

Loveman, Brian (2001) Chile: The Legacy of Hispanic Capitalism, 3rd

Edition, New York; Oxford University Press.

Malefakis, Edward E. (1970) Agrarian Reform and Peasant Revolution in

Spain, Yale University Press, New Haven CT.

Marglin, Stephen A. (1974) \What do Bosses Do?" Review of Radical

Political Economy, 6, 60-112.

Martins, Jos�e de Souza (1996) \Clientalism and Corruption in Contempo-

39

rary Brazil," in Walter Little and Eduardo Posada-Carb�o eds. Political Corruption

in Europe and Latin America, MacMillan, London.

Millar Carvacho, Ren�e (1981) La Elecci�on Presidencial de 1920: Ten-

dencias y Pr�acti